Electronic medical record systems (EMR) are held back because data entry is slow, expensive, and codified in a format unsuitable for relational database repositories. Medical Reporting Solutions has developed advanced technology, which can overcome these problems through new methods to encode knowledge in medical reports, and a domain specific speech recognition system capable of real-time natural language understanding. Our research and development uses methods in corpus linguistics and sentential logic to represent the knowledge in free-text medical reports in an efficient, codeable manner. We have created a prototype speech recognizer capable of natural language processing in real-time employing our unique knowledge base. The knowledge base, while under development, is derived from hundreds of thousands of reports in the radiology domain. Our project plan includes completing our radiology speech recognizer, substantially enlarging our semantic knowledge base to cover 60% of the domain, and extensively testing the system. We plan to test our recognizer by measuring the word error rate (WER), using three different recognition algorithms we believe will substantially enhance performance beyond the best commercial medical speech recognizers.